After almost 5 years that I started writing my Kafka client from scratch in elixir it finally have it's first complete release allowing users to produce and consume records using the most recent Kafka developments. Feedback is very welcome!
https://t.co/yaRMNXlaZd
@coproduto Pra variar, depnde. Mas uma heurística que gosto de usar é pensar:
"O código que está sendo gerado poderia ser uma library mantida por um terceiro?"
Se sim, o rigor é bem menor, e garantir input/output quase sempre é suficiente!
Se não, eu prefiro entender tudo no detalhe.
@johncrickett@Arvind_0136 I agree we don't have to, and we may push back this speed only focus.
But, if the focus is to learn it is hard to beat the "write it yourself" approach. LLMs are still very useful when you are doing so, but the workflow is very different from the current agentic stuff.
@Arvind_0136@johncrickett Totally agree, and it's even worse because when you build the wrong thing yourself you almost always learn exactly what is wrong with it, sometimes even before finishing it.
Not so sure this is true about agentic coding, because it strip away most of the learning!
@TheVixhal I find particularly interesting is the interplay between them. You often catch research problems during development.
I fear automating the development mode may lead to gaps in research.
Wrote about this here: https://t.co/0DzolLj4L5
Would love to know your thoughts on it!
@VictorTaelin Yes! Would love to know your thoughts on https://t.co/0DzolLj4L5
I've tried to organize my thoughts regarding this problem and came up with this text! (:
@stewartlynch8 That's exactly my concern! I've written this to organize my thoughts regarding this, would love to hear your feedback! https://t.co/0DzolLjCAD
@rfleury Exactly! And it is even worse, because even if LLMs generated the exact same code you would you missed the learning from writing it yourself.
I had the same feeling (that nobody is talking about), so I wrote this, would love to get your feedback: https://t.co/0DzolLj4L5
@burkov I'm skeptical about this, but even if it's true, any one who did a large scale system migration knows that the new code is the easiest part of it.
@rakyll My main concern is that if you did not spend the last year building it, you wouldn't be this "good judge" of the produced artifacts.
IMO, the byproduct of programming is not just the code, but the understanding, and nothing better to get that than writing the code by yourself!
@TheGingerBill It can be useful in a context with very explicit definitions of what can change between versions.
Because users can know if what they rely on may change with the new version. The main problem is nobody really knows everything they rely on. So yeah, kinda useless! hehe
@hnshah I call it "friction aversion".
We try to optimize things so much to be frictionless and effortless that we forget that without any friction you can't move at all.
IMHO, one of the most valuable skills is being able to identify good friction and don't avoid it.
@coproduto Já diria Dijkstra, em uma das minhas frases preferidas:
"Simplicity is a great virtue but it requires hard work to achieve it and education to appreciate it. And to make matters worse: complexity sells better."
@coproduto Em casos extremos, o logger da BEAM muda de async para sync automaticamente para evitar overload. https://t.co/RpwMbjSTQc
Nesses casos pode acabar impactando, já vi casos em prod de um sistema (em elixir) passar de 600 rps para mais de 1k rps apenas removendo logs.
@ChShersh You really are running out of time each second, so the feeling is accurate.
The beauty is to accept the fact that time being a finite resource is what makes your choices valuable.
If you had infinite time nothing would really matter, you could always do it later.
@GrugDev @guynikan Boa, não tinha entendido isso.
Mas ainda assim o ponto vale, você pode saber os motivos de antemão e ainda assim descobrir novas alternativas durante o desenvolvimento.
Não tô dizendo que tu tá certo ou errado, só tentando pensar junto nas causas da sua observação! (: